Interview
Luisa and Robert Long on how to make independent research more fun (2023)
- Plans to utilize probability distributions to express bimodal outcomes (very few or very many) rather than middle-ground assumptions, intending to input initial probabilities between zero and 99 and refine these into narrower ranges as the model develops.
- Anticipates that publishing will occur at a single, torturous moment triggered by hard deadlines and external pressure, aiming to publish preliminary results on complex topics like consciousness and nuclear war despite significant model uncertainty and imperfection.
- Commits to explaining reasoning behind probability inputs to facilitate critique and allows others to substitute different probabilities, arguing that transparency prevents the need for a "final certain word" and helps readers evaluate claims.
- Expects that future large language models will provide superior feedback and utility for research tasks compared to current versions, potentially serving as tools for brainstorming, day planning, and providing "HR-like" supportive feedback.
- Outlines a shift in personal strategy regarding intrinsic motivations, moving away from self-coercion or "bullying" resistant parts via rationality tools toward negotiating a "seat at the table" for interests that do not align directly with maximum impact.
- Identifies that perfectionism, anxiety, and imposter syndrome are primary barriers to finishing work, noting that few individuals in the rationality community can complete and publish without hard deadlines, payment, or accountability checks.
- Predicts that nuclear war modeling involves significantly more uncertainty than previous work, though initial inputs based on concrete data (such as smoke lofting) may yield specific numbers, with assumptions regarding city targeting likely increasing model nuance.
- Acknowledges limitations in personal focus, noting that a broad natural disposition makes maintaining month-to-month or quarter-to-quarter focus difficult compared to day-to-day tasks, leading to a reliance on external structures rather than internal willpower.
- Views the "uniform distribution" as a natural human instinct when uncertain and anticipates adding probability bumps to a baseline, while cautioning that completely solving complex philosophical problems like consciousness or AI is preposterous at the outset.
- References successful precedents in charity effectiveness research by noting that preliminary results from Ellie Hassenfeld and Holden Karnofsky were published and updated, demonstrating value in reporting current best guesses even if questions remain unsettled.
- Recognizes a failure mode in using cognitive behavioral therapy or rationality tools to suppress intrinsic motivations, warning that "locking away" core parts of oneself leads to eventual resistance and reduced long-term impact.
- Concludes that the gold standard for work-life integration is not constant happiness and frictionless impact, but rather finding the intersection between what is valuable for the world and what one intrinsically enjoys, acknowledging that impact may suffer from frequent job changes.
- Details specific operational constraints for content production, including reliance on external deadlines to overcome procrastination and the expectation that without these pressures, complex topics could be researched for years without publication.